The term “symbiosis” describes a mutually beneficial relationship between two parties that allows each to profit in tandem with the other. It is typically used to describe natural phenomena, such as the relationship between bees and plant pollen. But it is also an apt term to describe the way that machine learning-based AI and humans will develop and evolve together.
While 2017 was a hugely important year in terms of advancements in natural language processing, it was also a year of human frustration — and even mockery — over AI-based interactions. From Microsoft’s Zo (Tay’s new and improved successor) to the infamous BabyQ incident in China, we’ve seen that sometimes machines don’t learn quite fast enough for our human taste. But when we shut down technologies for not giving the right answer (as China did with BabyQ), we forget the very purpose of machine learning: to learn through human interaction. And while the impetus for teaching should not be placed so strongly as it has been on the consumer, we should take note of failures and teach machines to improve. In doing so, we will develop a mutually beneficial relationship wherein we help machines learn and they improve human efficiency.
Positive Feedback Loops Are Good For Business
Since the advent of machine learning in the late 1960s, programmers have struggled to develop robots that can learn without being explicitly programmed. The two primary issues they’ve faced include a dearth of training data and the difficulty of pattern identification. The past 10 years, however, have demonstrated the power of machine learning when these issues diminish: It has been used in speech and facial recognition, in warning systems such as predicting cyberattacks based on chatter and even in simple tasks like email filtering. 2017, in particular, marked a huge advancement in machine learning, when Google’s DeepMind division released its latest version of AlphaZero. The computer program rapidly reached Grandmaster level in chess after just four hours of playing, using a technique called reinforcement learning — an outcome that could conceivably be replicated in other applications.
When humans teach machines how to improve task performance over time, machines grow alongside humans in a symbiotic relationship. In the customer service industry, for instance, chatbots and response suggestion engines improve over time based on user interactions — both globally and individually. A machine becomes increasingly adroit at identifying user intent based on feedback and a steady stream of data.
In mathematical terms, this stimulated growth from two entities is called a positive feedback loop. A drives growth in B, which in turn drives growth in A. This type of relationship drives exponential, accelerated growth toward extreme values — an outcome that is not typically desirable in systems that require equilibrium (such as the body’s internal environment) but is highly desirable in a system that requires innovation. In the business world, companies that establish a positive feedback loop between employees and machine learning-based technologies will see accelerated growth.
To use the customer service example, as a chatbot becomes better and better at ticket deflection through suggesting knowledge-based articles, human agents have more time to tackle high-level issues, thereby decreasing backlog and customer dissatisfaction. As agents become increasingly skilled at tackling high-level issues, they can automate at higher and higher levels.